The National Highways Authority of India (NHAI) is using artificial intelligence (AI) to help officials navigate the organisation’s policies, contracts and highway project data, while working to preserve institutional knowledge that can be lost when officers are transferred.
Mayank Kumawat, an AI engineer at NHAI, and Deep Saikia, Deputy Manager (Information Technology) at the authority, outlined the initiative at the recent IndiaFOSS 2026, the annual conference organised by FOSS United. Their presentation described how NHAI’s internal AI team is using open-source software to build tools for everyday administrative and technical work, and how it is exploring ways to make employees’ knowledge reusable across the organisation.
Mr. Kumawat is part of NHAI’s AI Cell, where he helps build internal AI tools, including the authority’s assistant ‘Marg Sarthi’. According to his IndiaFOSS, he has worked on applied AI, education technology and public-sector systems, was a lead contributor to the Ministry of Statistics and Programme Implementation’s MCP initiative, and built an adaptive AI learning system called SMART in 2023 in collaboration with XLRI and Harvard faculty.
Deep Saikia (left), Deputy Manager (Information Technology), and Mayank Kumawat, an AI engineer at NHAI. | Photo Credit: Special arrangement
Mr. Saikia, a Computer Science and Engineering graduate from the National Institute of Technology (nit), Silchar, works on AI-based tools, digital transformation and workflow automation at NHAI. His work focuses on applying technology to public infrastructure and large-scale institutional processes.
Mr. Kumawat said that when he joined NHAI, he found that officials had to consult multiple sources to carry out routine tasks. These included laws, Ministry of Road Transport and Highways circulars, more than 1,800 internal NHAI policy circulars, and standards issued by the Indian Roads Congress. More complex tasks, such as approving contractor bills, drafting notices, handling land acquisition and obtaining environmental clearances, required officials to navigate several rules and documents.
He said officials often maintained their own folders of relevant circulars and guidelines, updating them as new documents were issued. But frequent transfers and deputations meant that incoming officers had to rebuild this knowledge base.
Mr. Saikia said the problem was not a lack of expertise within NHAI, but the difficulty of making that expertise reusable across the organisation. “The problem is that the knowledge is not always reusable,” he said during the presentation, adding that when officers move or are transferred, some of the context they have accumulated moves with them.
The team is experimenting with converting recurring administrative and technical processes into reusable workflows. Mr. Saikia said experienced officers could explain how particular processes work, identify the documents and approvals required, and have other experts review those procedures before they are incorporated into the system.
He said this could allow a workflow understood in one office to be used in another, and help prevent mistakes identified by one officer from being repeated elsewhere.
An assistant for everyday work
Mr. Kumawat demonstrated how ‘Marg Sarthi’, NHAI’s internal AI assistant, allows officers to query project data and policy documents through a conversational interface.
In one demonstration, Mr. Kumawat showed how an officer overseeing highway projects in Bengaluru could use the AI assistant to identify unresolved problems across projects. The system searched monthly progress reports, grouped the problems by category and displayed them on a dashboard. It could also generate a PDF report that the officer could send to contractors to seek explanations and follow up on the issues.
In another demonstration, the assistant was asked whether an additional work proposal worth ₹50 crore could be approved for a highway project originally valued at ₹250 crore. Rather than immediately answering yes or no, the system asked for more details about the contract and the officer’s authority to approve the proposal. It then provided an answer with references to the relevant rules, allowing the officer to check the original documents and verify the basis for its response.
The team said the system was designed to make answers auditable. During the question-and-answer session, the speakers distinguished between errors arising from retrieving incomplete or irrelevant information and errors in the answer generated from the retrieved material. They said citations and feedback from officers helped them identify and address gaps.
The speakers also said the assistant used role-based access controls linked to NHAI’s existing project-management system. An officer’s access through the assistant was intended to correspond to the projects and information that the officer was authorised to access in the underlying system.
NHAI to deploy 3D sensors to identify potholes, road cracks across 23 States
Open-source software
Mr. Kumawat said the team had built the assistant using existing open-source software rather than developing every component from scratch. LibreChat provides the conversational interface, while the Model Context Protocol (MCP) connects it to tools and data sources. The team also uses PostHog and Langfuse for AI observability, and Forgejo for code hosting.
Mr. Saikia said open-source software allowed the team to experiment with different solutions, adapt existing tools to NHAI’s requirements and obtain feedback from officers without having to begin a lengthy development process for every new idea.
He said the team could test a proposed tool with users, continue developing it if it proved useful, or abandon it if it did not address the problem. Conventional software development could be time-consuming, involving requirements gathering, procurement, vendor onboarding and development. Open-source software allowed the team to experiment with solutions more quickly.
The team said officers had begun suggesting new applications and reporting gaps in the assistant’s responses. Mr. Saikia said this interaction had also changed the relationship between the technology team and the engineers using the system, with officers contributing knowledge of their own workflows rather than remaining only users of the tools.
‘Marg Sarthi’ had reportedly crossed 15,000 conversations and 50,000 questions from more than 1,000 NHAI employees, with two in three users returning at least once a week. Mr. Kumawat separately said the system had answered more than two lakh queries since its launch on April 18, 2026.
The team also described developing specialised tools for tasks such as analysing technical schedules and geotechnical and road-safety reports. Mr. Saikia said not every problem needed to be solved by adding a workflow to the assistant; some required dedicated tools for repetitive technical checks before an expert exercised judgment.